Show More
@@ -1,1039 +1,1040 b'' | |||
|
1 | 1 | # revlogdeltas.py - Logic around delta computation for revlog |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2005-2007 Matt Mackall <mpm@selenic.com> |
|
4 | 4 | # Copyright 2018 Octobus <contact@octobus.net> |
|
5 | 5 | # |
|
6 | 6 | # This software may be used and distributed according to the terms of the |
|
7 | 7 | # GNU General Public License version 2 or any later version. |
|
8 | 8 | """Helper class to compute deltas stored inside revlogs""" |
|
9 | 9 | |
|
10 | 10 | from __future__ import absolute_import |
|
11 | 11 | |
|
12 | 12 | import collections |
|
13 | 13 | import struct |
|
14 | 14 | |
|
15 | 15 | # import stuff from node for others to import from revlog |
|
16 | 16 | from ..node import ( |
|
17 | 17 | nullrev, |
|
18 | 18 | ) |
|
19 | 19 | from ..i18n import _ |
|
20 | 20 | |
|
21 | 21 | from .constants import ( |
|
22 | 22 | REVIDX_ISCENSORED, |
|
23 | 23 | REVIDX_RAWTEXT_CHANGING_FLAGS, |
|
24 | 24 | ) |
|
25 | 25 | |
|
26 | 26 | from ..thirdparty import ( |
|
27 | 27 | attr, |
|
28 | 28 | ) |
|
29 | 29 | |
|
30 | 30 | from .. import ( |
|
31 | 31 | error, |
|
32 | 32 | mdiff, |
|
33 | 33 | util, |
|
34 | 34 | ) |
|
35 | 35 | |
|
36 | 36 | # maximum <delta-chain-data>/<revision-text-length> ratio |
|
37 | 37 | LIMIT_DELTA2TEXT = 2 |
|
38 | 38 | |
|
39 | 39 | class _testrevlog(object): |
|
40 | 40 | """minimalist fake revlog to use in doctests""" |
|
41 | 41 | |
|
42 | 42 | def __init__(self, data, density=0.5, mingap=0, snapshot=()): |
|
43 | 43 | """data is an list of revision payload boundaries""" |
|
44 | 44 | self._data = data |
|
45 | 45 | self._srdensitythreshold = density |
|
46 | 46 | self._srmingapsize = mingap |
|
47 | 47 | self._snapshot = set(snapshot) |
|
48 | 48 | self.index = None |
|
49 | 49 | |
|
50 | 50 | def start(self, rev): |
|
51 | 51 | if rev == nullrev: |
|
52 | 52 | return 0 |
|
53 | 53 | if rev == 0: |
|
54 | 54 | return 0 |
|
55 | 55 | return self._data[rev - 1] |
|
56 | 56 | |
|
57 | 57 | def end(self, rev): |
|
58 | 58 | if rev == nullrev: |
|
59 | 59 | return 0 |
|
60 | 60 | return self._data[rev] |
|
61 | 61 | |
|
62 | 62 | def length(self, rev): |
|
63 | 63 | return self.end(rev) - self.start(rev) |
|
64 | 64 | |
|
65 | 65 | def __len__(self): |
|
66 | 66 | return len(self._data) |
|
67 | 67 | |
|
68 | 68 | def issnapshot(self, rev): |
|
69 | 69 | if rev == nullrev: |
|
70 | 70 | return True |
|
71 | 71 | return rev in self._snapshot |
|
72 | 72 | |
|
73 | 73 | def slicechunk(revlog, revs, targetsize=None): |
|
74 | 74 | """slice revs to reduce the amount of unrelated data to be read from disk. |
|
75 | 75 | |
|
76 | 76 | ``revs`` is sliced into groups that should be read in one time. |
|
77 | 77 | Assume that revs are sorted. |
|
78 | 78 | |
|
79 | 79 | The initial chunk is sliced until the overall density (payload/chunks-span |
|
80 | 80 | ratio) is above `revlog._srdensitythreshold`. No gap smaller than |
|
81 | 81 | `revlog._srmingapsize` is skipped. |
|
82 | 82 | |
|
83 | 83 | If `targetsize` is set, no chunk larger than `targetsize` will be yield. |
|
84 | 84 | For consistency with other slicing choice, this limit won't go lower than |
|
85 | 85 | `revlog._srmingapsize`. |
|
86 | 86 | |
|
87 | 87 | If individual revisions chunk are larger than this limit, they will still |
|
88 | 88 | be raised individually. |
|
89 | 89 | |
|
90 | 90 | >>> data = [ |
|
91 | 91 | ... 5, #00 (5) |
|
92 | 92 | ... 10, #01 (5) |
|
93 | 93 | ... 12, #02 (2) |
|
94 | 94 | ... 12, #03 (empty) |
|
95 | 95 | ... 27, #04 (15) |
|
96 | 96 | ... 31, #05 (4) |
|
97 | 97 | ... 31, #06 (empty) |
|
98 | 98 | ... 42, #07 (11) |
|
99 | 99 | ... 47, #08 (5) |
|
100 | 100 | ... 47, #09 (empty) |
|
101 | 101 | ... 48, #10 (1) |
|
102 | 102 | ... 51, #11 (3) |
|
103 | 103 | ... 74, #12 (23) |
|
104 | 104 | ... 85, #13 (11) |
|
105 | 105 | ... 86, #14 (1) |
|
106 | 106 | ... 91, #15 (5) |
|
107 | 107 | ... ] |
|
108 | 108 | >>> revlog = _testrevlog(data, snapshot=range(16)) |
|
109 | 109 | |
|
110 | 110 | >>> list(slicechunk(revlog, list(range(16)))) |
|
111 | 111 | [[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]] |
|
112 | 112 | >>> list(slicechunk(revlog, [0, 15])) |
|
113 | 113 | [[0], [15]] |
|
114 | 114 | >>> list(slicechunk(revlog, [0, 11, 15])) |
|
115 | 115 | [[0], [11], [15]] |
|
116 | 116 | >>> list(slicechunk(revlog, [0, 11, 13, 15])) |
|
117 | 117 | [[0], [11, 13, 15]] |
|
118 | 118 | >>> list(slicechunk(revlog, [1, 2, 3, 5, 8, 10, 11, 14])) |
|
119 | 119 | [[1, 2], [5, 8, 10, 11], [14]] |
|
120 | 120 | |
|
121 | 121 | Slicing with a maximum chunk size |
|
122 | 122 | >>> list(slicechunk(revlog, [0, 11, 13, 15], targetsize=15)) |
|
123 | 123 | [[0], [11], [13], [15]] |
|
124 | 124 | >>> list(slicechunk(revlog, [0, 11, 13, 15], targetsize=20)) |
|
125 | 125 | [[0], [11], [13, 15]] |
|
126 | 126 | |
|
127 | 127 | Slicing involving nullrev |
|
128 | 128 | >>> list(slicechunk(revlog, [-1, 0, 11, 13, 15], targetsize=20)) |
|
129 | 129 | [[-1, 0], [11], [13, 15]] |
|
130 | 130 | >>> list(slicechunk(revlog, [-1, 13, 15], targetsize=5)) |
|
131 | 131 | [[-1], [13], [15]] |
|
132 | 132 | """ |
|
133 | 133 | if targetsize is not None: |
|
134 | 134 | targetsize = max(targetsize, revlog._srmingapsize) |
|
135 | 135 | # targetsize should not be specified when evaluating delta candidates: |
|
136 | 136 | # * targetsize is used to ensure we stay within specification when reading, |
|
137 | 137 | densityslicing = getattr(revlog.index, 'slicechunktodensity', None) |
|
138 | 138 | if densityslicing is None: |
|
139 | 139 | densityslicing = lambda x, y, z: _slicechunktodensity(revlog, x, y, z) |
|
140 | 140 | for chunk in densityslicing(revs, |
|
141 | 141 | revlog._srdensitythreshold, |
|
142 | 142 | revlog._srmingapsize): |
|
143 | 143 | for subchunk in _slicechunktosize(revlog, chunk, targetsize): |
|
144 | 144 | yield subchunk |
|
145 | 145 | |
|
146 | 146 | def _slicechunktosize(revlog, revs, targetsize=None): |
|
147 | 147 | """slice revs to match the target size |
|
148 | 148 | |
|
149 | 149 | This is intended to be used on chunk that density slicing selected by that |
|
150 | 150 | are still too large compared to the read garantee of revlog. This might |
|
151 | 151 | happens when "minimal gap size" interrupted the slicing or when chain are |
|
152 | 152 | built in a way that create large blocks next to each other. |
|
153 | 153 | |
|
154 | 154 | >>> data = [ |
|
155 | 155 | ... 3, #0 (3) |
|
156 | 156 | ... 5, #1 (2) |
|
157 | 157 | ... 6, #2 (1) |
|
158 | 158 | ... 8, #3 (2) |
|
159 | 159 | ... 8, #4 (empty) |
|
160 | 160 | ... 11, #5 (3) |
|
161 | 161 | ... 12, #6 (1) |
|
162 | 162 | ... 13, #7 (1) |
|
163 | 163 | ... 14, #8 (1) |
|
164 | 164 | ... ] |
|
165 | 165 | |
|
166 | 166 | == All snapshots cases == |
|
167 | 167 | >>> revlog = _testrevlog(data, snapshot=range(9)) |
|
168 | 168 | |
|
169 | 169 | Cases where chunk is already small enough |
|
170 | 170 | >>> list(_slicechunktosize(revlog, [0], 3)) |
|
171 | 171 | [[0]] |
|
172 | 172 | >>> list(_slicechunktosize(revlog, [6, 7], 3)) |
|
173 | 173 | [[6, 7]] |
|
174 | 174 | >>> list(_slicechunktosize(revlog, [0], None)) |
|
175 | 175 | [[0]] |
|
176 | 176 | >>> list(_slicechunktosize(revlog, [6, 7], None)) |
|
177 | 177 | [[6, 7]] |
|
178 | 178 | |
|
179 | 179 | cases where we need actual slicing |
|
180 | 180 | >>> list(_slicechunktosize(revlog, [0, 1], 3)) |
|
181 | 181 | [[0], [1]] |
|
182 | 182 | >>> list(_slicechunktosize(revlog, [1, 3], 3)) |
|
183 | 183 | [[1], [3]] |
|
184 | 184 | >>> list(_slicechunktosize(revlog, [1, 2, 3], 3)) |
|
185 | 185 | [[1, 2], [3]] |
|
186 | 186 | >>> list(_slicechunktosize(revlog, [3, 5], 3)) |
|
187 | 187 | [[3], [5]] |
|
188 | 188 | >>> list(_slicechunktosize(revlog, [3, 4, 5], 3)) |
|
189 | 189 | [[3], [5]] |
|
190 | 190 | >>> list(_slicechunktosize(revlog, [5, 6, 7, 8], 3)) |
|
191 | 191 | [[5], [6, 7, 8]] |
|
192 | 192 | >>> list(_slicechunktosize(revlog, [0, 1, 2, 3, 4, 5, 6, 7, 8], 3)) |
|
193 | 193 | [[0], [1, 2], [3], [5], [6, 7, 8]] |
|
194 | 194 | |
|
195 | 195 | Case with too large individual chunk (must return valid chunk) |
|
196 | 196 | >>> list(_slicechunktosize(revlog, [0, 1], 2)) |
|
197 | 197 | [[0], [1]] |
|
198 | 198 | >>> list(_slicechunktosize(revlog, [1, 3], 1)) |
|
199 | 199 | [[1], [3]] |
|
200 | 200 | >>> list(_slicechunktosize(revlog, [3, 4, 5], 2)) |
|
201 | 201 | [[3], [5]] |
|
202 | 202 | |
|
203 | 203 | == No Snapshot cases == |
|
204 | 204 | >>> revlog = _testrevlog(data) |
|
205 | 205 | |
|
206 | 206 | Cases where chunk is already small enough |
|
207 | 207 | >>> list(_slicechunktosize(revlog, [0], 3)) |
|
208 | 208 | [[0]] |
|
209 | 209 | >>> list(_slicechunktosize(revlog, [6, 7], 3)) |
|
210 | 210 | [[6, 7]] |
|
211 | 211 | >>> list(_slicechunktosize(revlog, [0], None)) |
|
212 | 212 | [[0]] |
|
213 | 213 | >>> list(_slicechunktosize(revlog, [6, 7], None)) |
|
214 | 214 | [[6, 7]] |
|
215 | 215 | |
|
216 | 216 | cases where we need actual slicing |
|
217 | 217 | >>> list(_slicechunktosize(revlog, [0, 1], 3)) |
|
218 | 218 | [[0], [1]] |
|
219 | 219 | >>> list(_slicechunktosize(revlog, [1, 3], 3)) |
|
220 | 220 | [[1], [3]] |
|
221 | 221 | >>> list(_slicechunktosize(revlog, [1, 2, 3], 3)) |
|
222 | 222 | [[1], [2, 3]] |
|
223 | 223 | >>> list(_slicechunktosize(revlog, [3, 5], 3)) |
|
224 | 224 | [[3], [5]] |
|
225 | 225 | >>> list(_slicechunktosize(revlog, [3, 4, 5], 3)) |
|
226 | 226 | [[3], [4, 5]] |
|
227 | 227 | >>> list(_slicechunktosize(revlog, [5, 6, 7, 8], 3)) |
|
228 | 228 | [[5], [6, 7, 8]] |
|
229 | 229 | >>> list(_slicechunktosize(revlog, [0, 1, 2, 3, 4, 5, 6, 7, 8], 3)) |
|
230 | 230 | [[0], [1, 2], [3], [5], [6, 7, 8]] |
|
231 | 231 | |
|
232 | 232 | Case with too large individual chunk (must return valid chunk) |
|
233 | 233 | >>> list(_slicechunktosize(revlog, [0, 1], 2)) |
|
234 | 234 | [[0], [1]] |
|
235 | 235 | >>> list(_slicechunktosize(revlog, [1, 3], 1)) |
|
236 | 236 | [[1], [3]] |
|
237 | 237 | >>> list(_slicechunktosize(revlog, [3, 4, 5], 2)) |
|
238 | 238 | [[3], [5]] |
|
239 | 239 | |
|
240 | 240 | == mixed case == |
|
241 | 241 | >>> revlog = _testrevlog(data, snapshot=[0, 1, 2]) |
|
242 | 242 | >>> list(_slicechunktosize(revlog, list(range(9)), 5)) |
|
243 | 243 | [[0, 1], [2], [3, 4, 5], [6, 7, 8]] |
|
244 | 244 | """ |
|
245 | 245 | assert targetsize is None or 0 <= targetsize |
|
246 | 246 | startdata = revlog.start(revs[0]) |
|
247 | 247 | enddata = revlog.end(revs[-1]) |
|
248 | 248 | fullspan = enddata - startdata |
|
249 | 249 | if targetsize is None or fullspan <= targetsize: |
|
250 | 250 | yield revs |
|
251 | 251 | return |
|
252 | 252 | |
|
253 | 253 | startrevidx = 0 |
|
254 | 254 | endrevidx = 1 |
|
255 | 255 | iterrevs = enumerate(revs) |
|
256 | 256 | next(iterrevs) # skip first rev. |
|
257 | 257 | # first step: get snapshots out of the way |
|
258 | 258 | for idx, r in iterrevs: |
|
259 | 259 | span = revlog.end(r) - startdata |
|
260 | 260 | snapshot = revlog.issnapshot(r) |
|
261 | 261 | if span <= targetsize and snapshot: |
|
262 | 262 | endrevidx = idx + 1 |
|
263 | 263 | else: |
|
264 | 264 | chunk = _trimchunk(revlog, revs, startrevidx, endrevidx) |
|
265 | 265 | if chunk: |
|
266 | 266 | yield chunk |
|
267 | 267 | startrevidx = idx |
|
268 | 268 | startdata = revlog.start(r) |
|
269 | 269 | endrevidx = idx + 1 |
|
270 | 270 | if not snapshot: |
|
271 | 271 | break |
|
272 | 272 | |
|
273 | 273 | # for the others, we use binary slicing to quickly converge toward valid |
|
274 | 274 | # chunks (otherwise, we might end up looking for start/end of many |
|
275 | 275 | # revisions). This logic is not looking for the perfect slicing point, it |
|
276 | 276 | # focuses on quickly converging toward valid chunks. |
|
277 | 277 | nbitem = len(revs) |
|
278 | 278 | while (enddata - startdata) > targetsize: |
|
279 | 279 | endrevidx = nbitem |
|
280 | 280 | if nbitem - startrevidx <= 1: |
|
281 | 281 | break # protect against individual chunk larger than limit |
|
282 | 282 | localenddata = revlog.end(revs[endrevidx - 1]) |
|
283 | 283 | span = localenddata - startdata |
|
284 | 284 | while span > targetsize: |
|
285 | 285 | if endrevidx - startrevidx <= 1: |
|
286 | 286 | break # protect against individual chunk larger than limit |
|
287 | 287 | endrevidx -= (endrevidx - startrevidx) // 2 |
|
288 | 288 | localenddata = revlog.end(revs[endrevidx - 1]) |
|
289 | 289 | span = localenddata - startdata |
|
290 | 290 | chunk = _trimchunk(revlog, revs, startrevidx, endrevidx) |
|
291 | 291 | if chunk: |
|
292 | 292 | yield chunk |
|
293 | 293 | startrevidx = endrevidx |
|
294 | 294 | startdata = revlog.start(revs[startrevidx]) |
|
295 | 295 | |
|
296 | 296 | chunk = _trimchunk(revlog, revs, startrevidx) |
|
297 | 297 | if chunk: |
|
298 | 298 | yield chunk |
|
299 | 299 | |
|
300 | 300 | def _slicechunktodensity(revlog, revs, targetdensity=0.5, |
|
301 | 301 | mingapsize=0): |
|
302 | 302 | """slice revs to reduce the amount of unrelated data to be read from disk. |
|
303 | 303 | |
|
304 | 304 | ``revs`` is sliced into groups that should be read in one time. |
|
305 | 305 | Assume that revs are sorted. |
|
306 | 306 | |
|
307 | 307 | The initial chunk is sliced until the overall density (payload/chunks-span |
|
308 | 308 | ratio) is above `targetdensity`. No gap smaller than `mingapsize` is |
|
309 | 309 | skipped. |
|
310 | 310 | |
|
311 | 311 | >>> revlog = _testrevlog([ |
|
312 | 312 | ... 5, #00 (5) |
|
313 | 313 | ... 10, #01 (5) |
|
314 | 314 | ... 12, #02 (2) |
|
315 | 315 | ... 12, #03 (empty) |
|
316 | 316 | ... 27, #04 (15) |
|
317 | 317 | ... 31, #05 (4) |
|
318 | 318 | ... 31, #06 (empty) |
|
319 | 319 | ... 42, #07 (11) |
|
320 | 320 | ... 47, #08 (5) |
|
321 | 321 | ... 47, #09 (empty) |
|
322 | 322 | ... 48, #10 (1) |
|
323 | 323 | ... 51, #11 (3) |
|
324 | 324 | ... 74, #12 (23) |
|
325 | 325 | ... 85, #13 (11) |
|
326 | 326 | ... 86, #14 (1) |
|
327 | 327 | ... 91, #15 (5) |
|
328 | 328 | ... ]) |
|
329 | 329 | |
|
330 | 330 | >>> list(_slicechunktodensity(revlog, list(range(16)))) |
|
331 | 331 | [[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]] |
|
332 | 332 | >>> list(_slicechunktodensity(revlog, [0, 15])) |
|
333 | 333 | [[0], [15]] |
|
334 | 334 | >>> list(_slicechunktodensity(revlog, [0, 11, 15])) |
|
335 | 335 | [[0], [11], [15]] |
|
336 | 336 | >>> list(_slicechunktodensity(revlog, [0, 11, 13, 15])) |
|
337 | 337 | [[0], [11, 13, 15]] |
|
338 | 338 | >>> list(_slicechunktodensity(revlog, [1, 2, 3, 5, 8, 10, 11, 14])) |
|
339 | 339 | [[1, 2], [5, 8, 10, 11], [14]] |
|
340 | 340 | >>> list(_slicechunktodensity(revlog, [1, 2, 3, 5, 8, 10, 11, 14], |
|
341 | 341 | ... mingapsize=20)) |
|
342 | 342 | [[1, 2, 3, 5, 8, 10, 11], [14]] |
|
343 | 343 | >>> list(_slicechunktodensity(revlog, [1, 2, 3, 5, 8, 10, 11, 14], |
|
344 | 344 | ... targetdensity=0.95)) |
|
345 | 345 | [[1, 2], [5], [8, 10, 11], [14]] |
|
346 | 346 | >>> list(_slicechunktodensity(revlog, [1, 2, 3, 5, 8, 10, 11, 14], |
|
347 | 347 | ... targetdensity=0.95, mingapsize=12)) |
|
348 | 348 | [[1, 2], [5, 8, 10, 11], [14]] |
|
349 | 349 | """ |
|
350 | 350 | start = revlog.start |
|
351 | 351 | length = revlog.length |
|
352 | 352 | |
|
353 | 353 | if len(revs) <= 1: |
|
354 | 354 | yield revs |
|
355 | 355 | return |
|
356 | 356 | |
|
357 | 357 | deltachainspan = segmentspan(revlog, revs) |
|
358 | 358 | |
|
359 | 359 | if deltachainspan < mingapsize: |
|
360 | 360 | yield revs |
|
361 | 361 | return |
|
362 | 362 | |
|
363 | 363 | readdata = deltachainspan |
|
364 | 364 | chainpayload = sum(length(r) for r in revs) |
|
365 | 365 | |
|
366 | 366 | if deltachainspan: |
|
367 | 367 | density = chainpayload / float(deltachainspan) |
|
368 | 368 | else: |
|
369 | 369 | density = 1.0 |
|
370 | 370 | |
|
371 | 371 | if density >= targetdensity: |
|
372 | 372 | yield revs |
|
373 | 373 | return |
|
374 | 374 | |
|
375 | 375 | # Store the gaps in a heap to have them sorted by decreasing size |
|
376 | 376 | gaps = [] |
|
377 | 377 | prevend = None |
|
378 | 378 | for i, rev in enumerate(revs): |
|
379 | 379 | revstart = start(rev) |
|
380 | 380 | revlen = length(rev) |
|
381 | 381 | |
|
382 | 382 | # Skip empty revisions to form larger holes |
|
383 | 383 | if revlen == 0: |
|
384 | 384 | continue |
|
385 | 385 | |
|
386 | 386 | if prevend is not None: |
|
387 | 387 | gapsize = revstart - prevend |
|
388 | 388 | # only consider holes that are large enough |
|
389 | 389 | if gapsize > mingapsize: |
|
390 | 390 | gaps.append((gapsize, i)) |
|
391 | 391 | |
|
392 | 392 | prevend = revstart + revlen |
|
393 | 393 | # sort the gaps to pop them from largest to small |
|
394 | 394 | gaps.sort() |
|
395 | 395 | |
|
396 | 396 | # Collect the indices of the largest holes until the density is acceptable |
|
397 | 397 | selected = [] |
|
398 | 398 | while gaps and density < targetdensity: |
|
399 | 399 | gapsize, gapidx = gaps.pop() |
|
400 | 400 | |
|
401 | 401 | selected.append(gapidx) |
|
402 | 402 | |
|
403 | 403 | # the gap sizes are stored as negatives to be sorted decreasingly |
|
404 | 404 | # by the heap |
|
405 | 405 | readdata -= gapsize |
|
406 | 406 | if readdata > 0: |
|
407 | 407 | density = chainpayload / float(readdata) |
|
408 | 408 | else: |
|
409 | 409 | density = 1.0 |
|
410 | 410 | selected.sort() |
|
411 | 411 | |
|
412 | 412 | # Cut the revs at collected indices |
|
413 | 413 | previdx = 0 |
|
414 | 414 | for idx in selected: |
|
415 | 415 | |
|
416 | 416 | chunk = _trimchunk(revlog, revs, previdx, idx) |
|
417 | 417 | if chunk: |
|
418 | 418 | yield chunk |
|
419 | 419 | |
|
420 | 420 | previdx = idx |
|
421 | 421 | |
|
422 | 422 | chunk = _trimchunk(revlog, revs, previdx) |
|
423 | 423 | if chunk: |
|
424 | 424 | yield chunk |
|
425 | 425 | |
|
426 | 426 | def _trimchunk(revlog, revs, startidx, endidx=None): |
|
427 | 427 | """returns revs[startidx:endidx] without empty trailing revs |
|
428 | 428 | |
|
429 | 429 | Doctest Setup |
|
430 | 430 | >>> revlog = _testrevlog([ |
|
431 | 431 | ... 5, #0 |
|
432 | 432 | ... 10, #1 |
|
433 | 433 | ... 12, #2 |
|
434 | 434 | ... 12, #3 (empty) |
|
435 | 435 | ... 17, #4 |
|
436 | 436 | ... 21, #5 |
|
437 | 437 | ... 21, #6 (empty) |
|
438 | 438 | ... ]) |
|
439 | 439 | |
|
440 | 440 | Contiguous cases: |
|
441 | 441 | >>> _trimchunk(revlog, [0, 1, 2, 3, 4, 5, 6], 0) |
|
442 | 442 | [0, 1, 2, 3, 4, 5] |
|
443 | 443 | >>> _trimchunk(revlog, [0, 1, 2, 3, 4, 5, 6], 0, 5) |
|
444 | 444 | [0, 1, 2, 3, 4] |
|
445 | 445 | >>> _trimchunk(revlog, [0, 1, 2, 3, 4, 5, 6], 0, 4) |
|
446 | 446 | [0, 1, 2] |
|
447 | 447 | >>> _trimchunk(revlog, [0, 1, 2, 3, 4, 5, 6], 2, 4) |
|
448 | 448 | [2] |
|
449 | 449 | >>> _trimchunk(revlog, [0, 1, 2, 3, 4, 5, 6], 3) |
|
450 | 450 | [3, 4, 5] |
|
451 | 451 | >>> _trimchunk(revlog, [0, 1, 2, 3, 4, 5, 6], 3, 5) |
|
452 | 452 | [3, 4] |
|
453 | 453 | |
|
454 | 454 | Discontiguous cases: |
|
455 | 455 | >>> _trimchunk(revlog, [1, 3, 5, 6], 0) |
|
456 | 456 | [1, 3, 5] |
|
457 | 457 | >>> _trimchunk(revlog, [1, 3, 5, 6], 0, 2) |
|
458 | 458 | [1] |
|
459 | 459 | >>> _trimchunk(revlog, [1, 3, 5, 6], 1, 3) |
|
460 | 460 | [3, 5] |
|
461 | 461 | >>> _trimchunk(revlog, [1, 3, 5, 6], 1) |
|
462 | 462 | [3, 5] |
|
463 | 463 | """ |
|
464 | 464 | length = revlog.length |
|
465 | 465 | |
|
466 | 466 | if endidx is None: |
|
467 | 467 | endidx = len(revs) |
|
468 | 468 | |
|
469 | 469 | # If we have a non-emtpy delta candidate, there are nothing to trim |
|
470 | 470 | if revs[endidx - 1] < len(revlog): |
|
471 | 471 | # Trim empty revs at the end, except the very first revision of a chain |
|
472 | 472 | while (endidx > 1 |
|
473 | 473 | and endidx > startidx |
|
474 | 474 | and length(revs[endidx - 1]) == 0): |
|
475 | 475 | endidx -= 1 |
|
476 | 476 | |
|
477 | 477 | return revs[startidx:endidx] |
|
478 | 478 | |
|
479 | 479 | def segmentspan(revlog, revs): |
|
480 | 480 | """Get the byte span of a segment of revisions |
|
481 | 481 | |
|
482 | 482 | revs is a sorted array of revision numbers |
|
483 | 483 | |
|
484 | 484 | >>> revlog = _testrevlog([ |
|
485 | 485 | ... 5, #0 |
|
486 | 486 | ... 10, #1 |
|
487 | 487 | ... 12, #2 |
|
488 | 488 | ... 12, #3 (empty) |
|
489 | 489 | ... 17, #4 |
|
490 | 490 | ... ]) |
|
491 | 491 | |
|
492 | 492 | >>> segmentspan(revlog, [0, 1, 2, 3, 4]) |
|
493 | 493 | 17 |
|
494 | 494 | >>> segmentspan(revlog, [0, 4]) |
|
495 | 495 | 17 |
|
496 | 496 | >>> segmentspan(revlog, [3, 4]) |
|
497 | 497 | 5 |
|
498 | 498 | >>> segmentspan(revlog, [1, 2, 3,]) |
|
499 | 499 | 7 |
|
500 | 500 | >>> segmentspan(revlog, [1, 3]) |
|
501 | 501 | 7 |
|
502 | 502 | """ |
|
503 | 503 | if not revs: |
|
504 | 504 | return 0 |
|
505 | 505 | end = revlog.end(revs[-1]) |
|
506 | 506 | return end - revlog.start(revs[0]) |
|
507 | 507 | |
|
508 | 508 | def _textfromdelta(fh, revlog, baserev, delta, p1, p2, flags, expectednode): |
|
509 | 509 | """build full text from a (base, delta) pair and other metadata""" |
|
510 | 510 | # special case deltas which replace entire base; no need to decode |
|
511 | 511 | # base revision. this neatly avoids censored bases, which throw when |
|
512 | 512 | # they're decoded. |
|
513 | 513 | hlen = struct.calcsize(">lll") |
|
514 | 514 | if delta[:hlen] == mdiff.replacediffheader(revlog.rawsize(baserev), |
|
515 | 515 | len(delta) - hlen): |
|
516 | 516 | fulltext = delta[hlen:] |
|
517 | 517 | else: |
|
518 | 518 | # deltabase is rawtext before changed by flag processors, which is |
|
519 | 519 | # equivalent to non-raw text |
|
520 | 520 | basetext = revlog.revision(baserev, _df=fh, raw=False) |
|
521 | 521 | fulltext = mdiff.patch(basetext, delta) |
|
522 | 522 | |
|
523 | 523 | try: |
|
524 | 524 | res = revlog._processflags(fulltext, flags, 'read', raw=True) |
|
525 | 525 | fulltext, validatehash = res |
|
526 | 526 | if validatehash: |
|
527 | 527 | revlog.checkhash(fulltext, expectednode, p1=p1, p2=p2) |
|
528 | 528 | if flags & REVIDX_ISCENSORED: |
|
529 | 529 | raise error.StorageError(_('node %s is not censored') % |
|
530 | 530 | expectednode) |
|
531 | 531 | except error.CensoredNodeError: |
|
532 | 532 | # must pass the censored index flag to add censored revisions |
|
533 | 533 | if not flags & REVIDX_ISCENSORED: |
|
534 | 534 | raise |
|
535 | 535 | return fulltext |
|
536 | 536 | |
|
537 | 537 | @attr.s(slots=True, frozen=True) |
|
538 | 538 | class _deltainfo(object): |
|
539 | 539 | distance = attr.ib() |
|
540 | 540 | deltalen = attr.ib() |
|
541 | 541 | data = attr.ib() |
|
542 | 542 | base = attr.ib() |
|
543 | 543 | chainbase = attr.ib() |
|
544 | 544 | chainlen = attr.ib() |
|
545 | 545 | compresseddeltalen = attr.ib() |
|
546 | 546 | snapshotdepth = attr.ib() |
|
547 | 547 | |
|
548 | 548 | def isgooddeltainfo(revlog, deltainfo, revinfo): |
|
549 | 549 | """Returns True if the given delta is good. Good means that it is within |
|
550 | 550 | the disk span, disk size, and chain length bounds that we know to be |
|
551 | 551 | performant.""" |
|
552 | 552 | if deltainfo is None: |
|
553 | 553 | return False |
|
554 | 554 | |
|
555 | 555 | # - 'deltainfo.distance' is the distance from the base revision -- |
|
556 | 556 | # bounding it limits the amount of I/O we need to do. |
|
557 | 557 | # - 'deltainfo.compresseddeltalen' is the sum of the total size of |
|
558 | 558 | # deltas we need to apply -- bounding it limits the amount of CPU |
|
559 | 559 | # we consume. |
|
560 | 560 | |
|
561 | 561 | textlen = revinfo.textlen |
|
562 | 562 | defaultmax = textlen * 4 |
|
563 | 563 | maxdist = revlog._maxdeltachainspan |
|
564 | 564 | if not maxdist: |
|
565 | 565 | maxdist = deltainfo.distance # ensure the conditional pass |
|
566 | 566 | maxdist = max(maxdist, defaultmax) |
|
567 | 567 | |
|
568 | 568 | # Bad delta from read span: |
|
569 | 569 | # |
|
570 | 570 | # If the span of data read is larger than the maximum allowed. |
|
571 | 571 | # |
|
572 | 572 | # In the sparse-revlog case, we rely on the associated "sparse reading" |
|
573 | 573 | # to avoid issue related to the span of data. In theory, it would be |
|
574 | 574 | # possible to build pathological revlog where delta pattern would lead |
|
575 | 575 | # to too many reads. However, they do not happen in practice at all. So |
|
576 | 576 | # we skip the span check entirely. |
|
577 | 577 | if not revlog._sparserevlog and maxdist < deltainfo.distance: |
|
578 | 578 | return False |
|
579 | 579 | |
|
580 | 580 | # Bad delta from new delta size: |
|
581 | 581 | # |
|
582 | 582 | # If the delta size is larger than the target text, storing the |
|
583 | 583 | # delta will be inefficient. |
|
584 | 584 | if textlen < deltainfo.deltalen: |
|
585 | 585 | return False |
|
586 | 586 | |
|
587 | 587 | # Bad delta from cumulated payload size: |
|
588 | 588 | # |
|
589 | 589 | # If the sum of delta get larger than K * target text length. |
|
590 | 590 | if textlen * LIMIT_DELTA2TEXT < deltainfo.compresseddeltalen: |
|
591 | 591 | return False |
|
592 | 592 | |
|
593 | 593 | # Bad delta from chain length: |
|
594 | 594 | # |
|
595 | 595 | # If the number of delta in the chain gets too high. |
|
596 | 596 | if (revlog._maxchainlen |
|
597 | 597 | and revlog._maxchainlen < deltainfo.chainlen): |
|
598 | 598 | return False |
|
599 | 599 | |
|
600 | 600 | # bad delta from intermediate snapshot size limit |
|
601 | 601 | # |
|
602 | 602 | # If an intermediate snapshot size is higher than the limit. The |
|
603 | 603 | # limit exist to prevent endless chain of intermediate delta to be |
|
604 | 604 | # created. |
|
605 | 605 | if (deltainfo.snapshotdepth is not None and |
|
606 | 606 | (textlen >> deltainfo.snapshotdepth) < deltainfo.deltalen): |
|
607 | 607 | return False |
|
608 | 608 | |
|
609 | 609 | # bad delta if new intermediate snapshot is larger than the previous |
|
610 | 610 | # snapshot |
|
611 | 611 | if (deltainfo.snapshotdepth |
|
612 | 612 | and revlog.length(deltainfo.base) < deltainfo.deltalen): |
|
613 | 613 | return False |
|
614 | 614 | |
|
615 | 615 | return True |
|
616 | 616 | |
|
617 | 617 | # If a revision's full text is that much bigger than a base candidate full |
|
618 | 618 | # text's, it is very unlikely that it will produce a valid delta. We no longer |
|
619 | 619 | # consider these candidates. |
|
620 | 620 | LIMIT_BASE2TEXT = 500 |
|
621 | 621 | |
|
622 | 622 | def _candidategroups(revlog, textlen, p1, p2, cachedelta): |
|
623 | 623 | """Provides group of revision to be tested as delta base |
|
624 | 624 | |
|
625 | 625 | This top level function focus on emitting groups with unique and worthwhile |
|
626 | 626 | content. See _raw_candidate_groups for details about the group order. |
|
627 | 627 | """ |
|
628 | 628 | # should we try to build a delta? |
|
629 | 629 | if not (len(revlog) and revlog._storedeltachains): |
|
630 | 630 | yield None |
|
631 | 631 | return |
|
632 | 632 | |
|
633 | 633 | deltalength = revlog.length |
|
634 | 634 | deltaparent = revlog.deltaparent |
|
635 | 635 | sparse = revlog._sparserevlog |
|
636 | 636 | good = None |
|
637 | 637 | |
|
638 | 638 | deltas_limit = textlen * LIMIT_DELTA2TEXT |
|
639 | 639 | |
|
640 | 640 | tested = {nullrev} |
|
641 | 641 | candidates = _refinedgroups(revlog, p1, p2, cachedelta) |
|
642 | 642 | while True: |
|
643 | 643 | temptative = candidates.send(good) |
|
644 | 644 | if temptative is None: |
|
645 | 645 | break |
|
646 | 646 | group = [] |
|
647 | 647 | for rev in temptative: |
|
648 | 648 | # skip over empty delta (no need to include them in a chain) |
|
649 | 649 | while (revlog._generaldelta |
|
650 | 650 | and not (rev == nullrev |
|
651 | 651 | or rev in tested |
|
652 | 652 | or deltalength(rev))): |
|
653 | 653 | tested.add(rev) |
|
654 | 654 | rev = deltaparent(rev) |
|
655 | 655 | # no need to try a delta against nullrev, this will be done as a |
|
656 | 656 | # last resort. |
|
657 | 657 | if rev == nullrev: |
|
658 | 658 | continue |
|
659 | 659 | # filter out revision we tested already |
|
660 | 660 | if rev in tested: |
|
661 | 661 | continue |
|
662 | 662 | tested.add(rev) |
|
663 | 663 | # filter out delta base that will never produce good delta |
|
664 | 664 | if deltas_limit < revlog.length(rev): |
|
665 | 665 | continue |
|
666 | 666 | if sparse and revlog.rawsize(rev) < (textlen // LIMIT_BASE2TEXT): |
|
667 | 667 | continue |
|
668 | 668 | # no delta for rawtext-changing revs (see "candelta" for why) |
|
669 | 669 | if revlog.flags(rev) & REVIDX_RAWTEXT_CHANGING_FLAGS: |
|
670 | 670 | continue |
|
671 | 671 | # If we reach here, we are about to build and test a delta. |
|
672 | 672 | # The delta building process will compute the chaininfo in all |
|
673 | 673 | # case, since that computation is cached, it is fine to access it |
|
674 | 674 | # here too. |
|
675 | 675 | chainlen, chainsize = revlog._chaininfo(rev) |
|
676 | 676 | # if chain will be too long, skip base |
|
677 | 677 | if revlog._maxchainlen and chainlen >= revlog._maxchainlen: |
|
678 | 678 | continue |
|
679 | 679 | # if chain already have too much data, skip base |
|
680 | 680 | if deltas_limit < chainsize: |
|
681 | 681 | continue |
|
682 | 682 | if sparse and revlog.upperboundcomp is not None: |
|
683 | 683 | maxcomp = revlog.upperboundcomp |
|
684 | 684 | basenotsnap = (p1, p2, nullrev) |
|
685 | 685 | if rev not in basenotsnap and revlog.issnapshot(rev): |
|
686 | 686 | snapshotdepth = revlog.snapshotdepth(rev) |
|
687 | 687 | # If text is significantly larger than the base, we can |
|
688 | 688 | # expect the resulting delta to be proportional to the size |
|
689 | 689 | # difference |
|
690 | 690 | revsize = revlog.rawsize(rev) |
|
691 | 691 | rawsizedistance = max(textlen - revsize, 0) |
|
692 | 692 | # use an estimate of the compression upper bound. |
|
693 | 693 | lowestrealisticdeltalen = rawsizedistance // maxcomp |
|
694 | 694 | |
|
695 | 695 | # check the absolute constraint on the delta size |
|
696 | 696 | snapshotlimit = textlen >> snapshotdepth |
|
697 | 697 | if snapshotlimit < lowestrealisticdeltalen: |
|
698 | 698 | # delta lower bound is larger than accepted upper bound |
|
699 | 699 | continue |
|
700 | 700 | |
|
701 | 701 | # check the relative constraint on the delta size |
|
702 | 702 | revlength = revlog.length(rev) |
|
703 | 703 | if revlength < lowestrealisticdeltalen: |
|
704 | 704 | # delta probable lower bound is larger than target base |
|
705 | 705 | continue |
|
706 | 706 | |
|
707 | 707 | group.append(rev) |
|
708 | 708 | if group: |
|
709 | 709 | # XXX: in the sparse revlog case, group can become large, |
|
710 | 710 | # impacting performances. Some bounding or slicing mecanism |
|
711 | 711 | # would help to reduce this impact. |
|
712 | 712 | good = yield tuple(group) |
|
713 | 713 | yield None |
|
714 | 714 | |
|
715 | 715 | def _findsnapshots(revlog, cache, start_rev): |
|
716 | 716 | """find snapshot from start_rev to tip""" |
|
717 | 717 | if util.safehasattr(revlog.index, 'findsnapshots'): |
|
718 | 718 | revlog.index.findsnapshots(cache, start_rev) |
|
719 | 719 | else: |
|
720 | 720 | deltaparent = revlog.deltaparent |
|
721 | 721 | issnapshot = revlog.issnapshot |
|
722 | 722 | for rev in revlog.revs(start_rev): |
|
723 | 723 | if issnapshot(rev): |
|
724 | 724 | cache[deltaparent(rev)].append(rev) |
|
725 | 725 | |
|
726 | 726 | def _refinedgroups(revlog, p1, p2, cachedelta): |
|
727 | 727 | good = None |
|
728 | 728 | # First we try to reuse a the delta contained in the bundle. |
|
729 | 729 | # (or from the source revlog) |
|
730 | 730 | # |
|
731 | 731 | # This logic only applies to general delta repositories and can be disabled |
|
732 | 732 | # through configuration. Disabling reuse source delta is useful when |
|
733 | 733 | # we want to make sure we recomputed "optimal" deltas. |
|
734 | 734 | if cachedelta and revlog._generaldelta and revlog._lazydeltabase: |
|
735 | 735 | # Assume what we received from the server is a good choice |
|
736 | 736 | # build delta will reuse the cache |
|
737 | 737 | good = yield (cachedelta[0],) |
|
738 | 738 | if good is not None: |
|
739 | 739 | yield None |
|
740 | 740 | return |
|
741 | 741 | snapshots = collections.defaultdict(list) |
|
742 | 742 | for candidates in _rawgroups(revlog, p1, p2, cachedelta, snapshots): |
|
743 | 743 | good = yield candidates |
|
744 | 744 | if good is not None: |
|
745 | 745 | break |
|
746 | 746 | |
|
747 | 747 | # If sparse revlog is enabled, we can try to refine the available deltas |
|
748 | 748 | if not revlog._sparserevlog: |
|
749 | 749 | yield None |
|
750 | 750 | return |
|
751 | 751 | |
|
752 | 752 | # if we have a refinable value, try to refine it |
|
753 | 753 | if good is not None and good not in (p1, p2) and revlog.issnapshot(good): |
|
754 | 754 | # refine snapshot down |
|
755 | 755 | previous = None |
|
756 | 756 | while previous != good: |
|
757 | 757 | previous = good |
|
758 | 758 | base = revlog.deltaparent(good) |
|
759 | 759 | if base == nullrev: |
|
760 | 760 | break |
|
761 | 761 | good = yield (base,) |
|
762 | 762 | # refine snapshot up |
|
763 | 763 | if not snapshots: |
|
764 | 764 | _findsnapshots(revlog, snapshots, good + 1) |
|
765 | 765 | previous = None |
|
766 | 766 | while good != previous: |
|
767 | 767 | previous = good |
|
768 | 768 | children = tuple(sorted(c for c in snapshots[good])) |
|
769 | 769 | good = yield children |
|
770 | 770 | |
|
771 | 771 | # we have found nothing |
|
772 | 772 | yield None |
|
773 | 773 | |
|
774 | 774 | def _rawgroups(revlog, p1, p2, cachedelta, snapshots=None): |
|
775 | 775 | """Provides group of revision to be tested as delta base |
|
776 | 776 | |
|
777 | 777 | This lower level function focus on emitting delta theorically interresting |
|
778 | 778 | without looking it any practical details. |
|
779 | 779 | |
|
780 | 780 | The group order aims at providing fast or small candidates first. |
|
781 | 781 | """ |
|
782 | 782 | gdelta = revlog._generaldelta |
|
783 | 783 | # gate sparse behind general-delta because of issue6056 |
|
784 | 784 | sparse = gdelta and revlog._sparserevlog |
|
785 | 785 | curr = len(revlog) |
|
786 | 786 | prev = curr - 1 |
|
787 | 787 | deltachain = lambda rev: revlog._deltachain(rev)[0] |
|
788 | 788 | |
|
789 | 789 | if gdelta: |
|
790 | 790 | # exclude already lazy tested base if any |
|
791 | 791 | parents = [p for p in (p1, p2) if p != nullrev] |
|
792 | 792 | |
|
793 | 793 | if not revlog._deltabothparents and len(parents) == 2: |
|
794 | 794 | parents.sort() |
|
795 | 795 | # To minimize the chance of having to build a fulltext, |
|
796 | 796 | # pick first whichever parent is closest to us (max rev) |
|
797 | 797 | yield (parents[1],) |
|
798 | 798 | # then the other one (min rev) if the first did not fit |
|
799 | 799 | yield (parents[0],) |
|
800 | 800 | elif len(parents) > 0: |
|
801 | 801 | # Test all parents (1 or 2), and keep the best candidate |
|
802 | 802 | yield parents |
|
803 | 803 | |
|
804 | 804 | if sparse and parents: |
|
805 | 805 | if snapshots is None: |
|
806 | 806 | # map: base-rev: snapshot-rev |
|
807 | 807 | snapshots = collections.defaultdict(list) |
|
808 | 808 | # See if we can use an existing snapshot in the parent chains to use as |
|
809 | 809 | # a base for a new intermediate-snapshot |
|
810 | 810 | # |
|
811 | 811 | # search for snapshot in parents delta chain |
|
812 | 812 | # map: snapshot-level: snapshot-rev |
|
813 | 813 | parents_snaps = collections.defaultdict(set) |
|
814 | 814 | candidate_chains = [deltachain(p) for p in parents] |
|
815 | 815 | for chain in candidate_chains: |
|
816 | 816 | for idx, s in enumerate(chain): |
|
817 | 817 | if not revlog.issnapshot(s): |
|
818 | 818 | break |
|
819 | 819 | parents_snaps[idx].add(s) |
|
820 | 820 | snapfloor = min(parents_snaps[0]) + 1 |
|
821 | 821 | _findsnapshots(revlog, snapshots, snapfloor) |
|
822 | 822 | # search for the highest "unrelated" revision |
|
823 | 823 | # |
|
824 | 824 | # Adding snapshots used by "unrelated" revision increase the odd we |
|
825 | 825 | # reuse an independant, yet better snapshot chain. |
|
826 | 826 | # |
|
827 | 827 | # XXX instead of building a set of revisions, we could lazily enumerate |
|
828 | 828 | # over the chains. That would be more efficient, however we stick to |
|
829 | 829 | # simple code for now. |
|
830 | 830 | all_revs = set() |
|
831 | 831 | for chain in candidate_chains: |
|
832 | 832 | all_revs.update(chain) |
|
833 | 833 | other = None |
|
834 | 834 | for r in revlog.revs(prev, snapfloor): |
|
835 | 835 | if r not in all_revs: |
|
836 | 836 | other = r |
|
837 | 837 | break |
|
838 | 838 | if other is not None: |
|
839 | 839 | # To avoid unfair competition, we won't use unrelated intermediate |
|
840 | 840 | # snapshot that are deeper than the ones from the parent delta |
|
841 | 841 | # chain. |
|
842 | 842 | max_depth = max(parents_snaps.keys()) |
|
843 | 843 | chain = deltachain(other) |
|
844 | 844 | for idx, s in enumerate(chain): |
|
845 | 845 | if s < snapfloor: |
|
846 | 846 | continue |
|
847 | 847 | if max_depth < idx: |
|
848 | 848 | break |
|
849 | 849 | if not revlog.issnapshot(s): |
|
850 | 850 | break |
|
851 | 851 | parents_snaps[idx].add(s) |
|
852 | 852 | # Test them as possible intermediate snapshot base |
|
853 | 853 | # We test them from highest to lowest level. High level one are more |
|
854 | 854 | # likely to result in small delta |
|
855 | 855 | floor = None |
|
856 | 856 | for idx, snaps in sorted(parents_snaps.items(), reverse=True): |
|
857 | 857 | siblings = set() |
|
858 | 858 | for s in snaps: |
|
859 | 859 | siblings.update(snapshots[s]) |
|
860 | 860 | # Before considering making a new intermediate snapshot, we check |
|
861 | 861 | # if an existing snapshot, children of base we consider, would be |
|
862 | 862 | # suitable. |
|
863 | 863 | # |
|
864 | 864 | # It give a change to reuse a delta chain "unrelated" to the |
|
865 | 865 | # current revision instead of starting our own. Without such |
|
866 | 866 | # re-use, topological branches would keep reopening new chains. |
|
867 | 867 | # Creating more and more snapshot as the repository grow. |
|
868 | 868 | |
|
869 | 869 | if floor is not None: |
|
870 | 870 | # We only do this for siblings created after the one in our |
|
871 | 871 | # parent's delta chain. Those created before has less chances |
|
872 | 872 | # to be valid base since our ancestors had to create a new |
|
873 | 873 | # snapshot. |
|
874 | 874 | siblings = [r for r in siblings if floor < r] |
|
875 | 875 | yield tuple(sorted(siblings)) |
|
876 | 876 | # then test the base from our parent's delta chain. |
|
877 | 877 | yield tuple(sorted(snaps)) |
|
878 | 878 | floor = min(snaps) |
|
879 | 879 | # No suitable base found in the parent chain, search if any full |
|
880 | 880 | # snapshots emitted since parent's base would be a suitable base for an |
|
881 | 881 | # intermediate snapshot. |
|
882 | 882 | # |
|
883 | 883 | # It give a chance to reuse a delta chain unrelated to the current |
|
884 | 884 | # revisions instead of starting our own. Without such re-use, |
|
885 | 885 | # topological branches would keep reopening new full chains. Creating |
|
886 | 886 | # more and more snapshot as the repository grow. |
|
887 | 887 | yield tuple(snapshots[nullrev]) |
|
888 | 888 | |
|
889 | 889 | if not sparse: |
|
890 | 890 | # other approach failed try against prev to hopefully save us a |
|
891 | 891 | # fulltext. |
|
892 | 892 | yield (prev,) |
|
893 | 893 | |
|
894 | 894 | class deltacomputer(object): |
|
895 | 895 | def __init__(self, revlog): |
|
896 | 896 | self.revlog = revlog |
|
897 | 897 | |
|
898 | 898 | def buildtext(self, revinfo, fh): |
|
899 | 899 | """Builds a fulltext version of a revision |
|
900 | 900 | |
|
901 | 901 | revinfo: _revisioninfo instance that contains all needed info |
|
902 | 902 | fh: file handle to either the .i or the .d revlog file, |
|
903 | 903 | depending on whether it is inlined or not |
|
904 | 904 | """ |
|
905 | 905 | btext = revinfo.btext |
|
906 | 906 | if btext[0] is not None: |
|
907 | 907 | return btext[0] |
|
908 | 908 | |
|
909 | 909 | revlog = self.revlog |
|
910 | 910 | cachedelta = revinfo.cachedelta |
|
911 | 911 | baserev = cachedelta[0] |
|
912 | 912 | delta = cachedelta[1] |
|
913 | 913 | |
|
914 | 914 | fulltext = btext[0] = _textfromdelta(fh, revlog, baserev, delta, |
|
915 | 915 | revinfo.p1, revinfo.p2, |
|
916 | 916 | revinfo.flags, revinfo.node) |
|
917 | 917 | return fulltext |
|
918 | 918 | |
|
919 | 919 | def _builddeltadiff(self, base, revinfo, fh): |
|
920 | 920 | revlog = self.revlog |
|
921 | 921 | t = self.buildtext(revinfo, fh) |
|
922 | 922 | if revlog.iscensored(base): |
|
923 | 923 | # deltas based on a censored revision must replace the |
|
924 | 924 | # full content in one patch, so delta works everywhere |
|
925 | 925 | header = mdiff.replacediffheader(revlog.rawsize(base), len(t)) |
|
926 | 926 | delta = header + t |
|
927 | 927 | else: |
|
928 | 928 | ptext = revlog.revision(base, _df=fh, raw=True) |
|
929 | 929 | delta = mdiff.textdiff(ptext, t) |
|
930 | 930 | |
|
931 | 931 | return delta |
|
932 | 932 | |
|
933 | 933 | def _builddeltainfo(self, revinfo, base, fh): |
|
934 | 934 | # can we use the cached delta? |
|
935 | 935 | revlog = self.revlog |
|
936 | 936 | chainbase = revlog.chainbase(base) |
|
937 | 937 | if revlog._generaldelta: |
|
938 | 938 | deltabase = base |
|
939 | 939 | else: |
|
940 | 940 | deltabase = chainbase |
|
941 | 941 | snapshotdepth = None |
|
942 | 942 | if revlog._sparserevlog and deltabase == nullrev: |
|
943 | 943 | snapshotdepth = 0 |
|
944 | 944 | elif revlog._sparserevlog and revlog.issnapshot(deltabase): |
|
945 | 945 | # A delta chain should always be one full snapshot, |
|
946 | 946 | # zero or more semi-snapshots, and zero or more deltas |
|
947 | 947 | p1, p2 = revlog.rev(revinfo.p1), revlog.rev(revinfo.p2) |
|
948 | 948 | if deltabase not in (p1, p2) and revlog.issnapshot(deltabase): |
|
949 | 949 | snapshotdepth = len(revlog._deltachain(deltabase)[0]) |
|
950 | 950 | delta = None |
|
951 | 951 | if revinfo.cachedelta: |
|
952 | 952 | cachebase, cachediff = revinfo.cachedelta |
|
953 | 953 | #check if the diff still apply |
|
954 | 954 | currentbase = cachebase |
|
955 | 955 | while (currentbase != nullrev |
|
956 | 956 | and currentbase != base |
|
957 | 957 | and self.revlog.length(currentbase) == 0): |
|
958 | 958 | currentbase = self.revlog.deltaparent(currentbase) |
|
959 | 959 | if self.revlog._lazydelta and currentbase == base: |
|
960 | 960 | delta = revinfo.cachedelta[1] |
|
961 | 961 | if delta is None: |
|
962 | 962 | delta = self._builddeltadiff(base, revinfo, fh) |
|
963 | 963 | header, data = revlog.compress(delta) |
|
964 | 964 | deltalen = len(header) + len(data) |
|
965 | 965 | offset = revlog.end(len(revlog) - 1) |
|
966 | 966 | dist = deltalen + offset - revlog.start(chainbase) |
|
967 | 967 | chainlen, compresseddeltalen = revlog._chaininfo(base) |
|
968 | 968 | chainlen += 1 |
|
969 | 969 | compresseddeltalen += deltalen |
|
970 | 970 | |
|
971 | 971 | return _deltainfo(dist, deltalen, (header, data), deltabase, |
|
972 | 972 | chainbase, chainlen, compresseddeltalen, |
|
973 | 973 | snapshotdepth) |
|
974 | 974 | |
|
975 | 975 | def _fullsnapshotinfo(self, fh, revinfo): |
|
976 | 976 | curr = len(self.revlog) |
|
977 | 977 | rawtext = self.buildtext(revinfo, fh) |
|
978 | 978 | data = self.revlog.compress(rawtext) |
|
979 | 979 | compresseddeltalen = deltalen = dist = len(data[1]) + len(data[0]) |
|
980 | 980 | deltabase = chainbase = curr |
|
981 | 981 | snapshotdepth = 0 |
|
982 | 982 | chainlen = 1 |
|
983 | 983 | |
|
984 | 984 | return _deltainfo(dist, deltalen, data, deltabase, |
|
985 | 985 | chainbase, chainlen, compresseddeltalen, |
|
986 | 986 | snapshotdepth) |
|
987 | 987 | |
|
988 | 988 | def finddeltainfo(self, revinfo, fh): |
|
989 | 989 | """Find an acceptable delta against a candidate revision |
|
990 | 990 | |
|
991 | 991 | revinfo: information about the revision (instance of _revisioninfo) |
|
992 | 992 | fh: file handle to either the .i or the .d revlog file, |
|
993 | 993 | depending on whether it is inlined or not |
|
994 | 994 | |
|
995 | 995 | Returns the first acceptable candidate revision, as ordered by |
|
996 | 996 | _candidategroups |
|
997 | 997 | |
|
998 | 998 | If no suitable deltabase is found, we return delta info for a full |
|
999 | 999 | snapshot. |
|
1000 | 1000 | """ |
|
1001 | 1001 | if not revinfo.textlen: |
|
1002 | 1002 | return self._fullsnapshotinfo(fh, revinfo) |
|
1003 | 1003 | |
|
1004 | 1004 | # no delta for flag processor revision (see "candelta" for why) |
|
1005 | 1005 | # not calling candelta since only one revision needs test, also to |
|
1006 | 1006 | # avoid overhead fetching flags again. |
|
1007 | 1007 | if revinfo.flags & REVIDX_RAWTEXT_CHANGING_FLAGS: |
|
1008 | 1008 | return self._fullsnapshotinfo(fh, revinfo) |
|
1009 | 1009 | |
|
1010 | 1010 | cachedelta = revinfo.cachedelta |
|
1011 | 1011 | p1 = revinfo.p1 |
|
1012 | 1012 | p2 = revinfo.p2 |
|
1013 | 1013 | revlog = self.revlog |
|
1014 | 1014 | |
|
1015 | 1015 | deltainfo = None |
|
1016 | 1016 | p1r, p2r = revlog.rev(p1), revlog.rev(p2) |
|
1017 | 1017 | groups = _candidategroups(self.revlog, revinfo.textlen, |
|
1018 | 1018 | p1r, p2r, cachedelta) |
|
1019 | 1019 | candidaterevs = next(groups) |
|
1020 | 1020 | while candidaterevs is not None: |
|
1021 | 1021 | nominateddeltas = [] |
|
1022 | 1022 | if deltainfo is not None: |
|
1023 | 1023 | # if we already found a good delta, |
|
1024 | 1024 | # challenge it against refined candidates |
|
1025 | 1025 | nominateddeltas.append(deltainfo) |
|
1026 | 1026 | for candidaterev in candidaterevs: |
|
1027 | 1027 | candidatedelta = self._builddeltainfo(revinfo, candidaterev, fh) |
|
1028 |
if i |
|
|
1029 |
|
|
|
1028 | if candidatedelta is not None: | |
|
1029 | if isgooddeltainfo(self.revlog, candidatedelta, revinfo): | |
|
1030 | nominateddeltas.append(candidatedelta) | |
|
1030 | 1031 | if nominateddeltas: |
|
1031 | 1032 | deltainfo = min(nominateddeltas, key=lambda x: x.deltalen) |
|
1032 | 1033 | if deltainfo is not None: |
|
1033 | 1034 | candidaterevs = groups.send(deltainfo.base) |
|
1034 | 1035 | else: |
|
1035 | 1036 | candidaterevs = next(groups) |
|
1036 | 1037 | |
|
1037 | 1038 | if deltainfo is None: |
|
1038 | 1039 | deltainfo = self._fullsnapshotinfo(fh, revinfo) |
|
1039 | 1040 | return deltainfo |
General Comments 0
You need to be logged in to leave comments.
Login now